Agisoft and Micasense RedEdge Processing

[Dvo]

Is Agisoft Photoscan Pro able to accurately corregister the individual channels of fixed-frame multi-lens cameras such as the micasense rededge or parrot sequioa? If so, any specific settings or workflow to follow for such cameras?

It is important for vegetation índex Calculations, for example, that the resulting multi-channel mosaic has all bands properly aligned in space. Even a few pixels displacement may produce incorrect índices.

I have found a number of posts on the forum where users have had problems processing missions, but none reporting success...

Dvoralai

[Alexey Pasumansky]

Hello Dvoralai,

Latest version 1.2.5 should work fine with Micasense cameras, but you need to follow simple recommendations on adding images: use Add Photos command and select all the images at once (shouldn't be any missing bands for any camera positions, so five images per camera position for RedEdge, for example), PhotoScan will scan the images and shortly after will propose different options of chunk creation, you need to select Mutlispectral Chunk option.
For such kind of imagery PhotoScan will assume that all images related to the individual bands were taken from the same camera position, but the orientation may be different as well as lens parameters. So for each camera you will find several "planes" in the Camera Calibration window.

[Dvo]

Hi Alexey

Thanks for your reply.

I would like to confirm that I inderstand you correctly. You state that
"PhotoScan will assume that all images related to the individual bands were taken from the same camera position, but the orientation may be different as well as lens parameters." yet strictly speaking the 5 bands are not taken from exactly the same position because of physical lens displacement, and not just a rotation. Can the lens parameters account for a relative translation of bands, the coregistration I enquiries about?

D

[Alexey Pasumansky]

Hello Dvoralai,

The approach implemented in the actual version of PhotoScan assumes that there's no shift between the sensors and in case of aerial surveys where the distance to the scanned surface is significant such assumption shouldn't induce considerable errors.

[Dvo]

Hello Alexey

Unfortunately, spatial coregistration errors of even as little as 1-2 pixels will lead to incorrect vegetation Index maps.

One could think of processing each band separately, and then doing a coregistration in other software; however, I suppose it is quite possible that individual bands would differ spatially if different lens parameters are used, also leading to errors.

Do you know of any plans in the (very) near future to introduce between-band spatial coregistration in Photoscan processing?

There is an increasing availability of such multiple lens cameras on the market (at least 4 at this time that I am aware of), being used for purposes of e.g. NDVI map creation.

regards,
Dvoralai

[Alexey Pasumansky]

Hello Dvoralai,

In case the flight is performed more or less on the same altitude the distance between the sensors can be compensated by additional rotation, so the error is such case should be under 1 pix.